Multi-Class Suicide Risk Prediction on Twitter Using Machine Learning Techniques

Abstract

The enormous growth of Social Networking Sites (SNS) resulted in more virtual engagement of people in the last decade. Amount of data generated through these SNS is enormous, allowing researchers to analyse this Big data. People share their opinions and thoughts related to any topic of interest. As suicide is one the leading cause of death worldwide, it has become a hot topic on which different researchers are working. The Covid19 further amplified the crisis due to social isolation which is the main risk factor for suicide. The problem has usually been analysed and dealt through a physiological point of view using Questionnaires and face to face settings but social stigma prevents its efficacy. In our research, we use well-known machine learning algorithms for multi-classification of Suicidal risk on social media so that individuals having high risk could be identified and counselled properly to save precious human lives. The data has been experimented through four popular machine learning algorithms: Logistic Regression, Multinomial Naïve Bayes, Support Vector Machine and Decision tree. The results generated are impressive with F1 Score ranging from 0.74 to 0.97. The Best performing algorithm was Decision tree that achieved an F-measure of 0.97, 0.94 and 0.96 for classifying suicidal text into three levels of concern.